Proto-Romance Syllable Structure

Naomi Cull
University of Calgary

1.0 Introduction
According to a traditional view (see Lindsay 1894; Muller 1929; Elcock I 960; Pei 1976),
the Romance languages derived from Classical Latin (CL) with an intermediate stage of Vulgar
Latin (VL), as represented in 1.1

(I) Old Latin

I
Classical Latin
I
Vulgar Latin
~
Romance Languages

However, there are several reasons to believe that this was not the correct progression.
First, there are certain features Classical Latin does not share with the Romance languages. For
example, in Classical Latin c1uus was invariable, yet at an earlier stage, Plautine Latin (c. 184
B.C.), cuius, -a, -um was an inflected adjective. Although the Romance languages also use kuiu,
-a whose' as an adjective, Hall (1950: 19) claims that this is a feature of Old Latin lost in Classical
Latin. Second, there arefeatures of Classical Latin which the Romance languages 'show no trace
of' (Hall 1983:5), such as the passive voice, the future tense and most non-finite forms.2 Third,
the Romance languages also have the adjectival suffix +arju- '(person, thing) connected with ... '
that Classical Latin avoided using (Hall 1983:5).3
Pulgram (1973) states that the reason there are common features in Vulgar Latin (which
Pulgram equates to Proto-Romance) and pre-Latin lies not in the fact that these features were
quiescent while the more prestigious Classical Latin was in use, but that in fact those features
shared by both Proto-Romance and pre-Latin were in continuous use from the old to the proto
stage. However, since these features were regarded as sub-standard they were not reflected in the
more highly esteemed Classical Latin.

1 Based on Hall (1950:24).

2 Some of these features may simply have been lost in the Romance languages.
3 Those features evident in the Romance languages but lacking in Classical Latin could also be language

internal developments.
35
 Another reason to disregard Classical Latin as the source of the Romance languages is
that Classical Latin was a literary form, a language of writing, not speech. As Hall (1974:177)
describes it, Classical Latin was an 'artificial' language not spoken by the masses but understood
by them. Because linguists try to describe speech as it is spoken, many have tried to equate Proto-
Romance (PR) with Vulgar Latin. This too is unsatisfactory as Vulgar Latin is almost identical
with a later stage of Romance, Proto-Italo-Westem Romance (Hall 1950:8). If Vulgar Latin was
the same as Proto-Romance, we would not be able to derive the eastern or southern Romance
languages which have features not traceable to Vulgar Latin. Therefore, it is necessary to push
Proto-Romance farther back than Vulgar Latin.4 Hall (1950:24; cf. Pulgram 1975:38) argues for
a tree similar to the one below.

(2) Pre-Latin

~
Classical Latin Proto-Romance
I
Romance Languages
.
From the branch of Proto-Romance the various Romance languages break off into subgroups and
eventually we arrive at the modem Romance languages.
As can be seen in this representation, Classical Latin and Proto-Romance are sisters,
Proto-Romance is not the daughter of Classical Latin. This type of representation can best explain
why certain features are found in Classical Latin but not in Proto-Romance and vice versa. Of
course, Proto-Romance is a reconstruction based on the comparison of the Romance languages
and dialects, as Hall (1974:1) emphasizes repeatedly. Therefore, allestations of this stage are not
forthcoming. The reliance on the written form however, has hindered historical linguistics in the
past. Ancient Sanskrit documents led Indo-Europeanists to believe for a long time that Sanskrit
and Proto-Inda-European were one and the same. This of course has been proven incorrect and
reminds us not to rely solely on the written word when doing reconstruction.
In this paper I am going to argue that Pre-Latin and Proto-Romance shared the same
syllable structure and that Classical Latin was innovative. I believe both Pre-Latin and Proto-
. Romance had the syllable structure VC$CV and that in the case of certain consonant clusters, this
shifted in Classical Latin to become V$CCV in order to improve the syllabic contact. Thus it is

4 Pulgram (1975:42) likewise refrains from using the tenn Vulgar Latin as an equivalent to Proto-Romance
because Vulgar Latin is often equated with post-classical Latin and 'vulgar', meaning non-standard speech,
existed prior to Classical Latin.

36
from this Proto-Romance and Pre-Latin syllable structure that the modem Romance languages are
derived and on which some important later phonological processes are based.
From Cla~sical Latin verse and accent assignment there is much evidence to support the
claim of V$CCV. For example, when a short vowel was followed by a plosive plus a liquid, the
first syllable was light, e.g., V$CCV, as in te11~$brae, pl1$tris and pi1$plus (Allen 1973: 137-138).
However, there also exists evidence of an earlier heterosyllabic syllabification of a short vowel
followed by a consonant and liquid, for example, an e in an open syllable became l, yet in int~gra
we have~. much like that in i1~c$ta, whereas in i11.f1$cit we get i. This indicates then that the
original structure was int~g$ra and not 1nt~$gra which would have produced 1nf1$gra .
If Pre-Latin was V$CCV we would have to argue that it became VC$CV in early Latin
and then returned to V$CCV once again in Classical Latin. However, it would be difficult to
motivate the change from V$CCV to VC$CV, but there is motivation for the change from VC$CV
to V$CCV in Classical Latin in syllable contact improvement, as we will see later.
The developments of Classical Latin and Proto-Romance as I believe them to exist are
shown below.

(3) Pre-Latin VC$CV

CL PR
V$CrV VC$rV
V$CIV VC$1V
V$CGV VC$GV (where C =plosive or fricative)

The various diachronic phonological processes discussed below which took place in
Italian, Portuguese, Romanian and Catalan will show that Proto-Romance was VC$CV.

2.0 Introduction to the Theory

There have been many difficulties in attempting to reconstruct a standard syllable
structure for Proto-Romance. For instance, in dividing a consonant cluster which occurs word-
medially, do we divide it as VC$CV or as V$CCV? Murray and Vennemann (1983) (hereafter M
& V) discussed the problem of Proto-Germanic syllable structure and concluded that the correct
reconstruction was VC$CV. Some similarities between Germanic and Romance were also
reviewed. It is on the basis of this paper, as well as Murray ( 1987) .and Vennemann ( 1988) and
evidence from four Romance languages, that I hope to find support for the Proto-Romance syllable
structure shown above. I will begin by first discussing the principles of the Preference Law theory
(M & V 1983; Vennemann 1988), the framework in which I will be working. Then I will give
evidence from Romance languages which strongly suggests that the correct Proto-Romance
syllable structure is VC$CV. Finally, I will discuss the works of other authors (Pensado 1989;
Salverda de Grave 1930) which raise questions about the proto-forms and also provide some
answers.

37
2.1 Preference Laws
The overview of the Preference Laws in this paper will be based on Vennemann (1988)
with further expansion from other sources. Although Vennemann presents a number of Preference
Laws, in this paper I will only offer those which are pertinent to the discussion at hand.
It is well known that syllable structure can help explain certain phonological phenomena
such as stress and tone assignment (see Halle and Vergnaud 1987). Syllable structures can also
provide the basis for certain sound changes, such as syllable-initial glide strengthening in Modem
Spanish (Pensado 1989:128). Vennemann (1988) explains sound changes by referring to the
syllable structure of a particular language and the universal Preference Laws for syllable
structure.s
What is meant by a Preference Law? As Vennemann (1988:1) puts it: 'These laws
specify the preferred syllable patterns of natural languages as well as determine the direction of
syllable structure change'. Preference implies a better/worse condition rather than a natural/
unnatural condition. Therefore, the nature of syllable structures is gradient, that is, some syllable
structures are more preferred than others. "'Xis the more preferred in terms of (a given parameter
of) syllable structure, the more Y', where X is a phonological pattern and Y a gradable property
of X". An improvement to a syllable structure is a syllable structure change, but a change that
worsens syllable structure is not a syllable structure change; that is, the change is not motivated
by syllable structure according to Vennemann, but is a change which affects syllable structure but
is motivated by a different parameter. As an example Vennemann (2) mentions that vowel
copations, such as syncope, must always worsen syllable structures, given that the preferred
syllable structure is CV. In a sequence of CVCVCV, for example, syncope could produce CCVCV
which would be a less preferred, therefore worse, syllable structure.
Diachronic improvements will generalize from the worst to the best syllable structures,
as implied in 4.6
(4) Diachronic Maxim
Linguistic change on a given parameter does not affect a language structure as long as
there exist structures in the language system that are less preferred in terms of the relevant
preference law.
Synchronically, a language will not contain structures that are less preferred without also
containing structures that are more preferred.
(5) Synchronic Maxim
A language system will in general not contain a structure on a given parameter without
containing those structures constructible with the means of the system that are more
preferred in terms of the relevant preference law.
However, changes which operate along different parameters may alter the system so that there is

s The Preference Laws are to be construed as universals, however, individual languages can develop their
own language-specific tendencies which sometimes conflict with universal laws and so 'unnatural'
structures sometimes arise. This is to be expected by virtue of the fact that dialects each have different
influences that shape it in various ways. Even so, all languages share certain basic features that tie them
together and it is those features to which universal laws can apply.
6 For an earlier attempt at explaining the chronology of diachronic changes see Foley ( 1977).

38
not always an even transition from less preferred to more preferred structures (4 ). Thus a Jang uage
syst.em is never 'perfect'.

2.1.1 Consonantal Strength

The phonetic correlates of each speech sound in a language can be placed on a
Consonantal Strength Scale. The strength is measured by the 'degree of deviation from
unimpeded (voiced) air flow' (Vennemann 1988:8). Segments on the scale are arranged

1
hierarchically from weakest to strongest, as represented in 6 (Murray 1987: 118).7

(6) glides
weak liquids
nasals
voiced fricatives
voiceless fricatives/voiced stops
strong voiceless stops

These segments can also be put on a numerical strength scale (Mand V:S 19; cf. Hooper
1976:206).8 .

(7) voiceless
voiced fricatives, voiceless
glides liquids nasals fricatives voiced stops stops

2 3 4 5 6 >
The usefulness of giving each class a numerical value will be made clear shortly.

2.1.2 The Head Law

(8) Head Law
A syllable head is the more preferred: (a) the closer the number of speech sounds in the
head is to one, (b) the greater the Consonantal Strength value ofits onset, and (c) the more
sharply the Consonantal Strength drops from the onset toward the Consonantal Strength
of the following syllable nucleus.

Part (a) of the Head Law says that the number of speech sounds will be reduced in a head
with more than one speech sound, as shown in the Pllli example in 9 where consonant deletion has
taken place.

7 While a Consonantal Strength scale is considered universal, there may be language-specific values given
to each class. That is, in one language voiceless fricatives may be considered stronger than voiced stops rather
than equal to them, as is represented in 6. As well, among a particular class certain segments may be
considered stronger than others, for instance, labials may be considered stronger than dentals or vice versa.
8 For further justification behind such a scale see Foley (1970); Hooper (1976); and Vennemann (1972).

39
(9) ll,fajn! > uanM 'knowledge'
A syllable with no head is also less preferred and so epenthesis is one means of remedying
this situation, as in Northern German.
( 10) ?alt for alt, cha$7os for cM$os
The most preferred head is one with the greatest consonantal strength according to (b)
of the Head Law. Therefore a weak syllable head is often strengthened as in 11.
(11) Lt. Januarius It. gennaio /d'J../9 'January'
Lt. uiuere It. yiyere 'to live'
Head Law (c) states that the greater the slope of the head toward the nucleus, the more
preferred the head. That is, the first memberof the head should have a greater consonantal strength
than the next member, if there is one, and the nucleus should be weaker still. For example, given
that /ti is stronger than /d/, a head containing $tr is more preferred than one with $dr since the
slope is greater when the voiceless plosive is the initial member of the head rather than its voiced
counterpart.
Classical Greek had syllable heads consisting of a nasal plus a liquid. As the drop from
a nasal to a liquid is very slight, this type of complex head is not preferred. To improve this
condition either syllable head strengthening or consonant epenthesis occurred.
(12) +mro-t6s > hrot6s (strengthenin..g) 'mortal man'
cf. +a-mrot6s > +am$ro$tos > am$hro$tos (epenthesis) 'immortal'
2.1.3 The Coda Law
(13) Coda Law
A syllable coda is the more preferred: (a) the smaller the number of speech sounds in the
coda, (b) the less the Consonantal Strength of its offset, and (c) the more sharply 1he
Consonantal Strength drops from the offset toward the Consonantal Strength of 1he
preceding syllable nucleus.10
Whereas Head Law (a) stated that the preferred number of speech sounds in the head is
one, the Coda Law (a) states that zero speech sounds is preferred in the coda. Elimination of
members of the coda takes place in Icelandic word-medially and word-finally.
(14) hest+ur he~+bak hefilil 'horse'
/hs$tYr/ /hs$bak/ /ht:ss/

9 I have tentatively placed the affricates with the plosives on the Consonantal Strength scale.
10
While there might appear to be a conflict between Coda law (b) and (c ), the conflict is only apparent. The
Coda Law states that preferentially, there should be no coda, but if there is one, it should be stronger than the
preceding nucleus.

40
 Because a weak offset is preferred, codas tend to weaken rather than strengthen as was
the case for syllable heads. Coda weakening is evident in some Spanish dialects.

(15) Sp. sala.s. > dial. salah11 'halls'

Coda Law (c) is the inverse to Head Law (c). That is, in a complex coda the strongest
element should be the final speech sound of the group and the slope should be greatest from the
final segment of the cluster toward the preceding speech sound toward the nucleus. This means
that a complex coda of rt$ would be preferable to rd$ or It$, etc. (Vennemann 1988:27).12

2.1.4 The Syllable Contact Law

The Syllable Contact Law (hereafter SCL) is the most crucial Preference Law with
regards to this paper.

(16) Syllable Contact LawB

A syllable contact A$B is the more preferred, the less the Consonantal Strength of the
offset A and the greater the consonantal Strength of the onset B; more precisely-the
greater the characteristic difference CS(B)-CS(A) between the Consonantal Strength of
B and that of A.

Recall that in 7 the phonetic correlates were given a numerical value. A syllable contact
of r$t would yield an equation of 6- 2 = 4, whereCS(B) = t= 6 and CS(A) = r= 2. If the contact
were t$r, the result would be -4, from 2 - 6. The larger the difference, the better the contact;
therefore, the contact of r$t is preferred over t$r since it yields a higher output. The German
example in 17 illustrates this.

(17) Wartha/var$ta/, but Tatra/ta$tra/

In the first example, Wartha, there is a good syllable contact of r$t. The contact in the second
example, Tatra would be bad if it was syllabified as t$r as we have just seen. However, it is
tautosyllabic which is quite acceptable since it satisfies Head Law (c) (Vennemann 1988:41).
Now that we have reviewed the Preference Laws as set out by Venncmann, we can
continue with the problem at hand; namely, Proto-Romance syllable structure with evidence from
the histories of four Romance languages.

3.0 Italian
Murray (1987) (following Foley 1977:32) presented a Consonantal Strength scale for
Romance based on the sound changes which have occurred diachronically in the Romance
languages. In this scale labials are considered stronger than velars and dentals. Foley (l 977) finds
evidence for this from the fact that intervocalic dentals and velars are deleted in Spanish and French

11 h patterns as a glide in terms of its consonantal strength.

12 Of the liquids r is weaker than l as we will see shortly.
13 Vennemann (1988) refers to this law merely as the Contact Law. I chose the title which appears most
commonly in the literature.

41
as shown in 18 and 19.
(18) Lt. regal em Sp. real 'royal'
Lt. creQ.o Sp. creo 'I believe'
but Lt. ha.here Sp. haher [aBer] 'to have'
(19) Lt. legere Fr. lire 'to read'
Lt. anflam Fr. amie 'friend'
Lt. creQ.ere Fr. croire 'to believe'
Lt. VItam Fr. vie 'life'
but Lt. ha.here Fr. ayoir 'to have'
Lt. nuam Fr. riye 'riverbank'
Among the liquids I is taken to be stronger than r . Evidence for this comes from
Romanian where intervocalic I weakens tor, and also Spanish word-final I and Andalusian
syllable-final I both weaken tor.
(20) Lt. solem Rom. soare 'sun'
Sp. local(e) > lugar 'place'
And. alma > arm a 'soul'
The Consonantal Strength scale for Romance as presented by Murray (1987: 119) is given
below.14
(21) a) weak Ir b) weak velars, dentals c) weak j u 1
r 2
strong 1 I 3
strong labials N 4
v 5
d g 6
b f 7
t k 8
strong p 9

14
Murray has assumed that b and/ are of equal strength according to the scale in 6 shown earlier.

42
3.1 Geminationis
Modem Standard Italian is well known for its phonemically long consonants. These long
consonants or geminates are usually represented orthographically in Italian, for example.fatto
'fact' versus/ato 'fate'. In the first word the sequence of <ti> is long but in the second word, the
<I> is short. These long consonants are not always represented orthographically, however, as we
can see in the wordfig/ia [fiUa] 'daughter'. If we can explain the historical motivation for the
development of these long consonants in Italian as a syllable contact improvement, we will have
partial evidence for the syllable structure of Proto-Romance. That is, later phonological
developments in Italian should help furnish proof for the earlier syllable structure of Proto-
Romance.
We can begin by first looking at gemination of consonants before).

(22) Lt. sauiat It. saimia 'he knows (subj.)'

Lt. ruhiam It. rohQia 'madder' (plant name)
Lt. fugiat It. fuggja 'he flees (subj.)'
Lt. vind!.!miam It. vendemmia 'grape harvest'
Lt. filiam- It. fig!ia [fiUa] 'daughter'
Lt. +i'>atium It. PUllO [puttsol 'stench'
Lt. braffiium It. bra~io [bratt o] 'arm'
Lt. raQ.ium It. raggio 'ray'
but
Lt. aream Pied. aira (via +arJa) 'threshing area'
Both the voiced and voiceless series of Latin plosives as well as I geminated before) but r did not.
Gemination of these plosives also occurred before the glide IJ, though less regularly for I and again
not at all for r.

15 The historical development of gemination in Italian is not equal to the synchronic gemination of Italian,
that is, diachronic gemination in Italian cannot be equated to Raddoppiamento sintallico (Nespor and Vogel
1986: 165-168 and 170-175; Saltarelli (1983) refers to this process as Rafforzamento). N & V describe the
type of gemination which takes place across word boundaries as a resyllabification process within the
phonological phrase which occurs when a short stressed vowel word-finally is followed by a word-initial
consonant. Through resyllabification the word-initial consonant also becomes the coda of the preceding
syllable. This cannot explain the historical evolution of geminates in Italian however, since in words such
asjiglio 'son', Lt. nllus, which have a long vowel preceding the word-internal consonant, there would be
no explanation for the resyllabification according to N & V's analysis. Saltarelli (21) tries to explain
diachronic gemination through his theory of synchronic gemination which is when the coda of a branching
rhyme isempty,itreassociates to the coda of the following onset creating am bisyllabic segments (19). Again,
this analysis does not work since he sees diachronic gemination as an ameliorative process subsequent to
consonant deletion. That is, once the coda of a syllable is deleted, leaving an unassociated C along the CV
tier, the coda reassociates to the right creating a geminate consonant. This cannot be generalized to all types
of historical gemination however, since gemination did not only take place as a result of consonant deletion.
Also, the example he uses, Lt. rupta, It. rotta is traditionally considered an assimilation process (Pei 1954:58),
not a coda deletion process.

43
(23) Lt. sau.tJI It. seimi 'I knew'
Lt. hab.IJI It. ebb.i 'I had'
Lt. ~11Jl It. potti 'I could'
Lt. cadLil It. caddi 'I fell'
Lt. vollJI It. vol!i 'I wanted'
but
Lt. doluit Olt. dolve 'he suffered'
Lt. paruit It. parve 'it seemed'
This gemination can be explained using the Preference Laws of Syllable Structure and
assumingheterosyll~ification of the consonant and the following glide, that is VC$JVand VC$11V.
The SCL given in 16 is required to explain the change from VC$JV to VC$QV.
The SCL states that a head with a higher consonantal value and a weaker coda in the
preceding syllable is mpre preferred than the opposite situation. By referring to the strength scale
presented in 21 and by using the equation in 16 we can see that the contact evaluation for a syllable
structure of Vt$JV is 1 - 8 = -7, where I is the consonantal strength value of) and 8 is the value of
t . In 22 and 23 Latin shows a poor syllable contact in that the weaker element, the glide, is in the
head position and the stronger element is in the coda which is the opposite of what is most preferred
according to the SCL. To rectify this, gemination occurred. That is, the consonant of the coda
duplicated so that the following head would be filled by a stronger identical consonant instead of
a glide thereby resulting in a more preferred syllable contact. The Proto-Romance syllable contact
of S$W (where S =strong and W = weak) became the more preferred S$S in modem Italian.
If syllable contact is actually the motivation behind gemination, why then do we see
gemination of I but not r when r is in the coda and the following syllable head is a glide?
Gemination of r does not occur because of the two liquids, r is the weaker one. The syllable
contact of r$J is more tolerable than 1$) since r is only slightly stronger than the following glide,
therefore there is less motivation to geminate.16 This tendency for the worst syllable contacts to
change before or instead of the better syllable contacts is explained in a corollary principle from
Murray (1987:121):
(24) The probability of a syllable structure undergoing a syllable structure improve-ment
process increases as its evaluation decreases.
Since the contact evaluation for 1$) is -2 and -1 for r$J, syllable contact change is most
likely to occur in the structure with the lower evaluation.
3.1.1 Slope Steepening
Slope steepening is a process by which the second element of syllable head weakens in
order to make the slope from the initial element in the cluster towards the proceeding syllable
nucleus steeper. That is, the consonantal strength decreases from the initial segment of a

16 Compare to West Germanic gemination in section i.t below.

44
consonantal cluster towards the following nucleus. For example, a sequence of $pl becoming $pr
or $p) would be an example of slope steepening. The Latin I became) in Italian when it was
preceded by a tautosyllabic plosive.

(25) Lt. p!ilnum It. piano 'floor'

Lt. p!ango It. piango 'I weep; I grieve'
Lt. c!amo It. chiamo 'I call'
Slope steepening can also be seen word internally:

(26) Lt. incladere It. inchiudere 'to shut in'

Lt. sarc(u)lum It. sarchio 'hoe'
Lt. templum It. tempio 'temple'
Lt. exemplum It. scempio 'havoc; ruin'
Lt. con flat It. gonfia 'swollen; inflated'

(27) Syllable Initial Margin Law

The preference for a syllabic structure $AB , where a and b are the consonantal
strength values of A and B respectively, increases with the value of a minus b.

Similar to Head Law (c) in 8, the Syllable Initial Margin Law states that a complex
syllable head should consist of a stronger initial speech sound followed by a weaker second speech
sound followed by an even weaker nucleus. The less a syllable head fulfills this requirement, the
more likely it is for an improvement to occur. A syllable head of $kJ (evaluation 7, where k = 8
and)= 1, therefore 8 - 1=7) is more preferred than $kr (evaluation 6), and $kr is more preferred
than $kl (evaluation 5) (Murray 1987: 120); thus a syllable structure improvement will occur in the
complex head of $kl prior to any improvement in either $kr or $k,J. This is exactly what we see
occurring in 25 and 26. That is,/ weakened to a glide after a syllable-initial plosive, thus improving
the slope of the head. We must return again briefly to gemination.
Earlier I examined word-internal gemination before the glides) and u. Gemination of
plosives also occurred word-medially before I in Italian.

(28) Lt. duv.lum It. dOIJ.JJ.iO 'double'

Lt. o~(u)lum It. ob_io 'eye'
Lt. sab.(u)lum It. su]:iliia 'chisel'
but
Lt. teg(u)lum It. teg!ia [teUa] 'baking pan'
Lt. aglo It. aljo 'to the'

In the first three examples in 28 gemination occurred with p, k and b, whiled and gin the last two
examples instead assimilated to the following consonant. This is directly in accordance with the
Diachronic Maxim. Since p, k and bare all stronger than d and g on the Consonantal Strength scale,

45
improvement starts with the worst structures.
To reiterate, the purpose of this paper is to provide evidence for the Proto-Romance
syllable structure, specifically whether an intervocalic consonant cluster should be divided as
VC$CV or V$CCV. The different developments that have been discussed here, gemination and
slope steepening, show us that the division must have been VC$CV.11
Since the weakening of I to a glide only occurred after a syllable-initial plosive we are
led to believe that the correct syllable structure in 28 is V$CIV. However, if we take as our starting
point the heterosyllabic structure of p$1, k$1 and b$1 we get a clearer picture of the processes that
were involved. First, a poor contact of VC$/V1B was ameliorated by gemination, which yielded
VC$CIV. An example of this intermediate stage can be seen in the example from dialectal
Abruzzese subbla 'chisel' (Rohlfs 1966:348). This provided the environment for I to become).
If we begin by assuming that the plosive and the liquid are tautosyllabic we can explain the slope
steepening but not the gemination. Ifon the other hand we start with a heterosyllabic structure both
the gemination and the slope steepening are explained satisfactorily.
Weakening of r after a plosive did not occur in Italian, for example, Lt. primum, It. primo
'first' (Murray: 125), because a plosive +r constitutes a fairly good syllable head and it is only the
worst ones that will undergo any improvement. Word-internal slope steepening also did not occur:
(29) Lt. supra It. sopra 'above'
Lt. petram It. pietra 'stone'
Lt. lacrimam It. lacrima 'tear'
but
Lt. cOp(u)lam It. coppia 'couple'
Lt. vet(u)lum It. vecchio 'old'
Lt. oc{u)lum It. occhio 'eye'
The fact that I weakened after a plosive {after gemination took place) but r did not
suggests that there was a differential syllabification in Italian; both T$1 (>T$TI) and $Tr co-
occurred. That is, Proto-Romance VT$rV > Italian V$TrV. It might seem peculiar that both
sequences of a plosive plus yod and a plosive plus a lateral liquid underwent gemination while a
plosive plus r underwent resyllabification. Why would a sequence with a segment of intermediate
consonantal strength (r) undergo a different process? The answer lies in the chronology of
gemination and resyllabification. At an early stage in the development of Italian the very worst
syllable contact, a plosive plus yod, underwent gemination in order improve the contact.
Following this the next worst contact, a plosive plus r, sustained a different but equally effective
procedure, resyllabification. Finally, any remaining contacts that were deemed undesirable were
once again improved upon.19 Thus the contact between a plosive plus a lateral, the least
undesirable syllable contact, underwent a second stage of gemination.20
17 The fact that p, k and b were heterosyllabic when preceding I implies that g and d also were heterosyllabic
in this environment given that the first three segments have higher consonantal strength values than d and
g, which implies thatifp, k andb were still heterosyllabic, d andg must have been as well given thatd and
g, which make better syllable contacts with $1, would not undergo any syllable contact improvements until
after the worse contacts, p$1, k$1 and b$/, had changed
18 C denotes any of the plosives p, k or b here.
19 This will be elaborated on later.
20 The change from t$/ > t$tl did not occur in Italian as$tl is an unacceptable syllable head in any Romance
language. As can be seen in It. vecchio, t first became k and then underwent gemination.
46
 The fact that the two stages of geminmion are interrupted by an alternate process is not
unusual. Newton ( 1972) calls this type of rule ordering 'intcrdigitation'. In his paper the author
is referring to two stages of obstruent deletion which are separated by a stage of glide formation
in French. He states that 'what is intuitively a single phonoligica1 process is split into two parts
by a second rule, so that we may symbolize the situation as A BA '(41). Thus the two stages of
gemination suggested for Italian are plausible.

3.2 Unnatural Syllabifications

Pensado (1989) explains the 'unnatural' plosive plus glide syllabifications in Vulgar
Latin, (VC$CV) by means of juncture and metrics in Classical Latin. That is, she secs the Romance
languages as a continuation of Vulgar Latin which is itself a direct descendant of Classical Latin.
I have already stated that I disagree with this supposed evolution of the Romance languages.
Although I agree with her syllabification of VC$GV, I do not agree with her argumentation for this
syllabification; namely, that the 'unnatural' syllabic structure of VC$CV (in terms of the SCL) in
Vulgar Latin is due to a resyllabification of V$CCV in Classical Latin (Pensado: 133).
Pcnsado (132) concurs with Nyman's (1982) theory of lcxicalization which states that
lento speech constraints acted upon unnatural syllabic structures in Classical Latin containing
glides which evolved through allegro speech rules. That is, syllabic structure constraints of lento
speech created hcterosyllabic sequences of a consonant plus glide even though these types of
hcterosyllabic sequences had previously only been morphologically determined in Classical
Latin. For example, resyllabification did not occur across morpheme boundaries in a sequence of
a final vowel plus initialmll/a cum liquida even though this would create a better syllable structure,
e.g., CL lib /egione would be syllabified b$1 and not $bl; that is, the plosive and liquid were
syllabified morphologically despite the fact that a less preferred syllable structure would be the
result.
However, Pensado disagrees with Nyman who concludes that this type of rcsyllabification,
i.e., $CG> C$G, is unnatural and characteristic of poetic language in Classical Latin. Pcnsado
(133) instead would claim that this resyllahification was more general and in fact was the impetus
for allowing glides to occur word- and syllable-initially in Vulgar Latin where they had only
occurred syllabic-initially in Classical Latin and not morpheme-initially. Resyllabitication
according to Pcnsado relaxed the restriction against the distribution of g Iides. Thus Pensado ( 133)
would argue that 'lexicalizationcontradicting phonological naturalness is in fact possible' and that
while there is a tendency for natural rules to generalize. 'there seems also to be an opposite
tendency to preserve or even to generalize phonologically unnatural domain restrictions ( 134).
On the other hand, if we begin with VC$CV in Old Latin as I have suggested and continue
this sequence into Proto-Romance we do not have to accoum for the unnatural reversal of the
syllabic boundary in Vulgar Latin. While Pensado must explain how V$CCVbecame VC$CV in
Vulgar Latin, which she herself admits is an aberrant event, we merely have to explain how Proto-
Romance VC$CV became V$CCV in Classical Latin. which wi::, can do by referring to a natural
process of resyllahification due to poor syllable contact. Therefore, starting with VC$CV in the
proto-language simplifies the argument and docs not require rcfcral to unnatural phonological
processes.

47
3.3 Differential Developments in Italian
SalverdadeGrave(1930) (hereafter S deG) examines syllable structure in Latin and how
different types of syllable structures explain the distinct developments of seemingly identical
sequences of speech sounds. He does not accept Grandgent's ( 1927) claim that these develop-
ments are due to dialect influences or that they arise through analogy. Instead, S de G claims that
these diverse developments are due to differences in syllabification (323), though he does not
explain how these different syllable divisions arise.
S de G (323) posits three types of syllabifications for intervocalic conson:mt clusters:
(30) I. The cluster stays together and
a. the consonant cluster is tautosyllabic, as in It. orecchio from Lt. auri$da 'ear'.
b. the consonant cluster forms one single sound and is 'intervocalic' e.g. It. origlia
also from Lt. auric/a.
II. The cluster is heterosyllabic, for example, the ng in pian$go 'I cry'.
Of course in this paper I assume that the clusters of which S de G speaks are all originally
heterosyllabic.
S de G's argument is weak on a number of points. First, in la of 30 he argues that in
orecchio the k and I must have been tautosyllabic in order to account for the slope steepening (/
>J), yet he does not explain how the geminate arose in this form. Second, in lb of 30 he claims
the plosive and liquid are a single sound. He offers no description of how these two segments are
syllabified, he merely states that they are intervocalic. Tuts gives us no indication as to why
origlia, with the geminate palatal lateral, should arise from auric/a.
The heterosyllabic theory I have presented so far, however, can explain both the
development of origlia and orecchio if we accept that the diversity in their developments can be
attributed to dialect differences.21
(31) a. auricla b. auricla
k$1 k$1
g$1 Voicing k$kl Gemination
J$1 Coda Weakening k$kJ Slope Steepening
j$,\ Palatalization
,{$,{ Assimilation orecchio

As 31 shows there were two different developments based on the same initial syllable
structure. In the example on the left, origlia, the plosive first became voiced then weakened in the
coda position since it was followed by a weaker syllable head. Then the liquid following the yod
palatalized. Finally, the yod assimilated to the palatal lateral. In the second development,
orecchio, the regular development of k$1 proceeded, that is, gemination followed by slope

21 Rohlfs ( 1966:350) also attributes this difference lo dialectal variation.

48
steepening. The reason there were two developments could be attributed to dialect variation
(much like Grandgent theorized) with the possibility that minimal pairs existed, auriculum, where
syncope occurred after the voicing of the stop, and auric/um, in which there was no voicing and
the development proceeded as described above (cf. Pei 1954:63).
3.3.1 Consonant + glide
I would like to show some of the different developments of consonant cluster sequences
referred to in 30 and give S de G's explanation for their development and then my own.
(32) pJ la. It. sappia Lt. sa$pJam 'I know (pres. subj.)'
II It. saccia Lt. sap$Jam 'I know (pres. subj.)'
S de G explains the dialectal differences in 32 in the following manner: He maintains that
in II in 32 the syllable-initial yod in the heterosyllabic sequence underwent glide strengthening,
a common pro,s:essin Italian, fotexample Lt. major em> It. maggiore [ddf] 'greater; elder'. Instead
of the usual <fL though we get r5 in saccia since the yod follows a voiceless consonant. As for the
p before the glide in II, it is in a weak syllable-final position and so is lost according to S de G.
However, he does not explain why gemination, i.e., <cc> has occurred. As for la in 32, with the
tautosyllabic -pJ-, S de G states that p andJ are so different in articulation that the pis maintained
(324). S de G gives the same types of reasoning for differential developments involving bJ. mi.
vJ and others.22 However, there is another theory which can more coherently explain the diverse
developments.
We must first assume, as I have throughout this paper, that the sequence of intervocalic
consonant+ glide is heterosyllabic. Both of the results shown in 32 can be derived from this kind
of structure, as represented in 33.
(33) a. sapiam b. sapiam
p$J z p$J
p$d Glide Strengthening p$pJ Gemination
R$t~ Voicing Assimilation
t~$t~ A'ssimilation sappia
saccia
The difference in the developments may once again be due to dialect variation where it
is possible that the etymon of saccia maintained its trisyllabic state longer than that of sappia. The
developments in 33 also explain why there is a voicing distinction between maggiore and saccia,
since in the latter form there was a voicing assimilation, whereas in the former there was only glide
strengthening with subsequent assimilation. Since both developments can be explained with one
type of syllabification, there is no need to posit another as S de G has done, for which he has no
explanation.

22See 324-327 for further examples and similar explanations from S de G on intervocalic consonant+ glide
developments.

49
3.3.2 Muta cum Iiquida
Salverda de Grave argues that a consonant plus a liquid were never hcterosyllabified
word-internally (327). That this is not the case I think has been demonstrated earlier in this paper,
at least for/. In 3.1.1, the section on slope steepening, I said that there was a difference in Italian
in the syllabification of word-medial I and rafter they followed a plosive, namely T$1 and $Tr.
However, C$r must have been present at some point in Proto-Romance. How else are we to
explain Italian words with gemination before r such asfebbro from Lt.fabrum 'blacksmith' (S de
G~327)?
As I explained in 3.1.l ,resyllabification did take place afteran earlier stage of gemination
transpired. If we start with a structure such ac; T$J, the worst syllable contact, we know a bad
syllable contact exists. One way to correct this is gemination as we saw earlier. Another way to
improve a bad contact is resyllabification. In Italian resyllabification of the.word-internal plosive
did not occur when preceding I but did occur before r because a syllable-initial plosive plus I is
less preferred than a syllable-initial plosive plus r according to the Syllable Initial Margin Law.
Therefore, after the initial stage of gemination of a plosive preceding yod, resyllabification of the
word-internal heterosyllabic plosive occurred when preceding r. When resyllabification occurred
with r the first segments to resyllabify would be the voiceless plosives according to the Diachronic
Maxim. After this stage of resy llabification another phase of gemination took place. In this second
stage of gemination all the remaining syllable contacts considered undesirable were eliminated
starting with the worst' sequence p$1 and continuing down the Consonantal Strength scale to b$r.
This sequence of events can explain the cases of gemination and resyllabification in Italian without
making reference to differential syllabifications which S de G relies on. A summary of the changes
1just described is shown below with the changes listed in the order they occurred.
Table 1

Proto-Romance Vp$J Vb$J Vp$rV Vb$rV Vp$1 Vb$!

Vp$pJ
__ .., _____
Vb$bJ -------- -------- -------- -------- Gemination I
-------- V$prV -------- -------- -------- Resyllabification
-------- -------- -------- Vb$brV Vp$pl Vb$bl Gemination 2
-------- -------- -------- -------- Vp$pJ Vb$bJ Slope Steepening
Italian Vp$pJ Vb$bJ V$prV Vb$brV Vp$pJ Vb$bJ

In this section I examined the processes of gemination and slope-steepening in Italian and
how they show that a word-internal consonant cluster must have been divided as VC $CV. The
development of 'unnatural' syllable structures in Vulgar Latin as presented by Pensa do ( 1989) was
briefly reviewed and it was concluded that her results were at least inconclusive. I also reviewed
work from Salverdade Grave (1930) to show that only one type of syllabification is necessary to
explain all of the phonological developments in Italian. I hope to find corroborating evidence in
my further examination of 0th.er Romance languages.

50
4.0 Portuguese
Evidence from another Romance language, Portuguese, also alludes to an original
VC$CV sequence. I would like to extend the principles referred to in the section on Italian to the
Portuguese data below in order to show that the phonological processes of metathesis, coda
weakening, and glide strengthening present in the historical development of Standard Portuguese
support the theory that Proto-Romance did indeed have the syllable structure VC$CV.

4.1 Metathesis23
In Italian recall the most common means of ameliorating a poor syllable contact was
gemination. In Portuguese, however, the most typical remedy was metathesis, as shown in 34.

(34) VL se~iam Pg. siba24 'cuttlefish'

VL sapiat Pg. saiba2s 'know (pres. subj.)'
VL pil~um Pg. p~o26 'well'
VL Mdie Pg. hoje27 'today'
VL +rabiam Pg. raiva 'rabies, anger'
VL n1~um Pg. ruivo 'ruddy, reddish'
CL cav~am Pg. gaiva '!
VL basium Pg. beijois 'kiss'
VL bassiare Pg. baixar 'to lower'
VL querimoii'iam Pg. caramunha '?
VL vindemtam Pg. vindima 'grape harvest'
VL pecalium Pg. pegulho29 '?
CL januaiium Pg. janeiro30 'January'
CL diaiia Pg. ge!ra 'a yoke of land'
VL cl'.Siium Pg. cmro 'leather, hide'
VL contraiium Pg. contrairo (old and popular) 'contrary'

23 Data for the historical development of Portuguese comes from Williams (1962) which I will be
abbreviating as Wms.
24 VL tonic e + j > j or ei .
25 The form aipo 'celery' cf. CL lipt11m is slightly irregular in that we expect intervocalic voicing of the p
as is the case in saiba . It is possible aipo derived from +appiu according to Williams (79) in which case
we could posit degemination (which was prevalent in Portuguese, e.g. Lt. parm11m, Pg. pano 'cloth' ). This
degemination would create another sequence of p$J which would result in metathesis and produce j$p. If the
intervocalic voicing preceded degemination then we would expect to have two different developments, p >
b and pp> p. This seems to be the case judging by the development of geminates in Portuguese, e.g. CL
abbatem, Pg. abade 'abbot', CL cipp11m, Pg. cepo 'log; stump', CL biiccam, Pg. boca 'mouth' .
26 VL tonic o + j > o, 11, oi, or 11i
27 VL tonic :> + j > o or oi .
28 Also with palatalization of the s .
29 VL tonic 11 + j > 11 .
30 VL Lonie a + j > ai or ei .

51
 ' In siba, for example, the derivation would have been the following: sep$jam > srb$jam
(voicing could have occurred al this point or at some point later in the derivation) > seJ$ba
(metathesis)> siba.
(35) sl!p$iam
p$J
b$J Intervocalic Voicing
J$b Metathesis
si$ba Vowel Coalescence
siba
Metathesis also occurred to a lesser extent with 11.
(36) VL sapilit Pg. soube31 'I/he knew (pret.)'
VL ~tilit Pg. p6de32 'he/she could (pret. indic.)'
VL placilit Pg. prougue ?
VL habait Pg. houve 'I/he had (pret.)'
VL ~quam 6qua > dial. euga 'mare'
VL ~silit Pg. p6s 'he/she put (pret. indic.)'
The development of houve would be hab$11it > hav$11it (frication of the b) > ha11$ve >
houve [ove].
As shown in 34 all of the consonants which underwent metathesis when followed by a
heterosyllabic yod have a higher consonantal strength value than the following yod (refer to the
Consonantal strength scale 21). Therefore, it is predicted by the Syllable Contact Law that a
contact.of a strong coda followed by a weaker head will be improved. By reversing the positions
of the coda and the head, we now have a weaker coda than head which is more preferred.
In some examples in 34 the process of metathesis is concealed due to palatalization, for
example, Portugues~por;o, hoje, beijo and baixar. However, we know that metathesis did indeed
occur due to the change in the vowel originally preceding the consonant + yod sequence. By
investigating the development of certain vowels when followed by yod, we can see that metathesis
must have taken place in order to explain how the vowel and the yod came into contact.33
The fact that k and g appear not to have undergone metathesis when followed by a yod,
as indicated by the lack of change in the preceding vowel, for example Lt.facie, Pg.face 'face',
Lt. pu/egium , Pg. poejo 'pennyroyal mint', suggests that two parameters were at work,
palatalization, an assimilatory process, on the one hand, and metathesis, a syllable contact
improvement process, on the other. If palatalization was to apply prior to metathesis, it would in
effect bleed the environment for metathesis in some instances. Thus those segments completely
palatalized prior to metathesis would not undergo metathesis. It appears from the data that the
velars were first in line for palatalization. This is not unexpected given that velars often undergo

31 VL tonic a+ II >ou [o].

32 VL tonic ~ + 11 > o .
33 The footnotes beside the examples in 43 and 44 show the developments of the vowels in question.

52
palatalization before the labials and dentals (Foley 1977:94). It seems that the two parameters
were in competition with each other. At an early stage palatalization was productive and affected
the velars first. Prior to metathesis the velars had undergone the full process of palatalization. The
dentals on the other hand seem to have undergone only partial palatalization before the second
parameter, metathesis, took effect. Thus those segments which had already undergone complete
palatalization, k and g, were prevented from also undergoing metathesis, but those sounds which
had undergone only partial palatalization were still in the environment, S$W, for metathesis to
occur. This is represented below.

(37)

Palatalization Metathesis
k p

g d b

What 37 shows is that at the point at which metathesis becomes productive, the velars had
sustained the entire process of palatalization but the dentals had only partially undergone
palatalization. Thus the dentals as well as the labials were in the environment for metathesis.

fac$je bas$jum
(38) Stage 1: Intervocalic voicing baz$jum
Stage 2: Complete Palatalization of the velars: fa$ce
Stage 3: Partial Palatalization of the dentals. baf$jo
Stage 4: Metathesis of the dentals and labials. - - b;:U$!o

face beijo

What does all of this mean for our reconstruction? It provides further support for an
earlier syllabification of VC$CV and not V$CCV. While it appears in the examples above with the
velars before the yod that the plosives could have been tautosyllabic with the glide, it is more likely
that heterosyllabification was the true structure if we accept that the other plosives were also
heterosyllabic. If the consonant and the glide in those instances of partial palatalization had been
tautosyllabic there would have been no justification for the metathesis since a complex head
containing a consonant and a glide is quite acceptable. However, a syllable structure of a strong
coda followed by a weaker head is not acceptable, as we have seen, therefore it needs to be
improved. If the dental and labial plosives were heterosyllabic, the velars must have been as well

53
given that a language system will not contain a less preferred structure (p$j) without also
containing a more preferred structure (k$J). This was the case in Proto-Romance and metathesis
in Portuguese supports this conclusion.
4.1.l Coda Weakening
In Italian a word-internal sequence of an intervocalic plosive plus a heterosyllabic, lateral
liquid first geminated and then slope steepening occurred. This did not occur in Portuguese
however. Instead, the plosive in the coda weakened when preceding the /. Recall that coda
weakening is another means of improving the syllable contact.
(39) VL aptc(ll)lam Pg. abelha34 'honey bee'
VL tsc(ll)lum Pg. 6lho 'eye'
VL rOt(ll)lum +roclam > Pg. r6lha3s 'cork'
VL teg(ll)lam Pg. tel ha 'tile'
VL fab{ll)lare fablar > fallar > Pg. falar36 'to speak'
(40) ap1c(ll)lam
apic$lam Vowel Deletion
abic$lam Intervocalic Voicing
abU$lam Coda Weakening
abi$.<a Palatalization
abe$.< a Misc. Vowel Changes
abelha
In 39 the plosives preceding I all underwent weakening. The b infaba/~re assimilated
to the following /. The development of intervocalic consonants differed from those presented
above.

34 VL k > J when p~eceded by a vowel and followed by I. This yod had the effect of palatalizing the lateral.
When the k was preceded by another consonant, cl instead became clz, e.g. VL mascrllum >Pg. macho 'male'
35 ti> kl in Vulgar Latin.
36 Words where bl> br, such as VL obligllre > obrigar 'to force', are borrowings from Spanish or are semi-
leamed. Thus the apparent tautosyllabification of bl (as evidenced by the slope steepening) is not the regular
syllabification but a later development. Williams (78) does not accept previous arguments to explain the
development of bl infalar, arguments such asfa/ar developed under the influence of calar or that it is due
to the dissimilation of v of the intermediate form +favlar. According to Williams these arguments do not
explain forms such as taleira 'bracket'(< +1abularia ), or ullo(< +ublo< r1b1illum) and solo 'solo'(<+ sublo
< srlb ilium). If however, we assume that the plosive and the lateral were heterosyllabic and that the plosive
underwent assimilation and then deletion, these words are also explained.

54
(41) CL lilpum Pg. lobo 'wolf'
CL am1cum Pg. amigo 'friend'
CL natam Pg. nada [l))37 'nothing'
CL legumen Pg. legume 'bean'
CL habe re Pg. haver 'to have'

We can also compare the changes in 39 to those of an intervocalic plosive plus r.

(42) VL apt'ilem Pg. abril 'April'

VL ~tram Pg. pedra 'stone'
VL laciimam Pg. lagrima 'tear'
VL ~brem Pg. fevreJs (old and pop.) 'fever'
but
VL inU!grum Pg. inteiro39 'whole, entire'

Unless we take as our starting point VC$CV there is no way of explaining this weakening.
If the plosives in 39 were indeed tautosyllabic with the following l we would expect developments
similar to those in 41 and 42; that is, intervocalic voicing or frication, but not coda weakening. This
leads to the next point of differential syllabification, which we briefly discussed in the section on
Italian.
Once again it appears that when a plosive precedes r it is tautosyllabic rather than
heterosyllabic. When a plosive precedes a liquid there is a preference for that plosive to be in either
the coda or head position, depending on the strength of the plosive and the strength of the liquid.
This preference is based on the Syllable Contact Law and the Syllable Initial Margin Law.

(43) Contact Evaluation Initial Margin Evaluation.j(]

p$r -7 $gl 3
p$1 -6 $gr 4
k$r -6 $dl 3
k$1 -5 $dr 441
t$r -6 $bl 4
t$1 -5 $br 5

37 t > d > IJ.

38 The fonn fragua 'furnace' (< +fra11ga < +fravg a <fravega <Lt.fabricam) shows not only the intervocalic
frication of b > v but also shows metathesis of r. What it also shows is the metathesis of the 11. that is, +frauga
> frag11a. Given that the form +Jra11ga seems to represent an ideal syllable contact, a weak coda followed
by a strong onset, there appears to be no justification for this metathesis. The answer could lie in the fact that
this word is a borrowing from Spanish and thus does not represent the regular development of this sequence
in Portuguese.
39 Where gr remains gr or where cl and g/become gr, these words are semi-learned or borrowings according
to Williams (77).
40 Based on Murray (1987:128).
41 Even though d and g have the same consonantal strength, g as an onset is less preferred given that velars
arc i11l11:rcntly weaker than dentals (Foley 1977:33).

55
 b$r -5 $ti 5
b$1 -4 $tr 6
d$r -4 $kl 5
d$1 -3 $kr 6
g$r -4 $pl 6
g$1 -3 $pr 7
Both the contact and initial margin columns show the least preferred sequences starting
from the top and going downward toward the most preferred sequences. What this table shows
is that r and a preceding plosive are preferentially tautosyllabic while I plus a preceding plosive
are preferentially heterosyllabic. That is, when one compares a sequence of a plosive plus r to a
sequence of a plosive plus/, tautosyllabic plosive plus r is preferred to tautosyllabic plosive plus
I and heterosyllabic plosive plus I is preferred to heterosyllabic plosive plus r.
This table also shows that of all the plosives, g makes the worst syllable head. Thus it
is not entirely unexpected that it remains in the coda position in int~grum while the other plosives
appear in the head position. That is, if the voiced velar plosive was to resyllabify and become
tautosyllabic with the following r, a relatively poor syllable head would be the result. The g being
in the coda position would explain why we get coda weakening as opposed to frication in inteiro.
The Portuguese example dobrar 'to duplicate', Lt. dupl!re appears to be an exception to
those forms in 39 since it seems to exhibit word-internal slope steepening which would indicate
tautosyllabification instead of heterosyllabification. However, if we compare this form to Pg.
doble 'double', Lt. diJplus, we see a different pattern. In the second example, doble, Proto-
Romance stress existed on the syllable preceding the voiceless plosive. As stressed syllables tend
to attract segments, this syllable was most likely closed by the plosive, e.g. fC$CV. Since dobrar
originally had stress following the plosive and liquid combination, the likely syllable structure was
V$CCf. Thus the syllabification and development of these two words was dependent upon stress
as shown in the derivation below.
(44) a. dOp$1us b. du$ptAre
du$prar Slope Steepening
dub$1u du$brar Voicing
dob$1e do$brar Misc. Vowel Changes
dobJe42 dobrar

4.1.2 Glide Strengthening

Yet another means of improving a sequence of VC$CV where the coda is stronger than
the head is head strengthening. In this case, we are talking about glide strengthening since it is the
glides which appear in the head position. The example in 45 shows the strengthening of 11 to v when
it occurs as a head.

42 The regular develpment of bl > II >I, as infalar, would have occurred at a stage prior to voicing, therefore
we do not get the development *duh$le > *dul$le > *dole in Portuguese.

56
(45) Lt. valyisset Pg. valvesse '?'
Lt. +dolyerunt Pg. dolyeron'?'
Lt. man.u_ale Pg. mangual 'flail'
Lt. +min,U.are Pg. minguar 'to decrease'
Because the consonantal strength of the glide is weaker than the preceding coda, the
glide's consonantal strength increases by becoming v in the first two examples and gu in the last
two.
I have again tried to show in this section how phonological processes in Portuguese, such
as metathesis, coda weakening, and glide strengthening, can give us an indication as to earlier
syllable structure. It seems apparent, judging by the evidence put forth, that there is strong reason
to believe that Proto-Romance had heterosyllabic intervocalic consonant clusters based on the
data from Portuguese and Italian. In the next section I will present data from Romanian which
should provide further proof for this hypothesis.

5.0 Romanian
5.1 Slope Steepening
(46) Lt. glanda Rom. ghincm 'acorn'
Lt. clave Rom. cheie 'key'
Lt. plaga Rom. plagli 'wound'
Lt. blasphemo NRom. blestema 'to curse'
Lt. flore- Rom. inflori(re) 'to tlower'
Word-initially only the velar plosives + l underwent slope steepening. Why should this
be so? Recall 43 in the Portuguese section where $pl made a better complex head than either $gl
or $kl. This was because p had a greater consonantal strength than either g or k (p = 9, k = 8, g =
6) and so is more tolerable as the initial segment of a complex head. That is, the slope of $pl is
6, $kl is 5 and $gl is 3. Therefore, the structures with the worst values, here $kl and $gl, will be
altered prior to more preferred structures, here $pl, according to the Diachronic Maxim.
However, two segments which lie between k and g on the consonantal strength scale,
band/, do not undergo this slope steepening as expected. Perhaps this can be explained by the
fact that velars are typically weak segments and have a tendency to weaken prior to labials:13 Thus
we get slope steepening first with the velars -kl- and -gl-, then if the phonological process is still
productive, it will carry on with the dentals then labials. As 47 shows, slope steepening stopped
after the velars. ,

(47) '
Slope Steepening
k p

g d

Word-internally slope steepening also took place, again only with the velars.

43 This fact is represented in 21 of the Italian section.

57
(48) Lt. oricla Rom. ureche 'ear'
Lt. vigilare Rom. veghea 'to watch'
Lt. vetulu- Rom. vcchiu 44 'old'
but
Lt. duplus Rom. duplu 'double'
and
Lt. stab(u)lu Rom. staul 'stable'
Lt. sub(u)lu Rom. sulll 'awl'
Rom. p<?vidlll colloq. povirlll ?

Word-internally we see a variety of different developments. In the first two examples

in 48 with -kl- and -gl- the plosive remains while the I is weakened toJ, then deleted. With -pl-
both the plosive and the lateral remain. Finally, with -bl- the plosive is deleted while the I remains.
One possible solution to these differential developments is resyllabification and coda weakening.
The Romanian forms ureche and veghea seem to suggest tautosyllabification since the
liquid after the plosive weakens to yod, thereby palatalizing the preceding velar, and then ulti-
mately disappearing. However, if we begin with a syllable structure of VC$CV (this would involve
prior vowel syncope for some of the forms in 48) followed by resyllabification, we end up with
a structure of V$CCV. Resyllabification would start with the least preferred structures and work
its way through the system. In Romanian resyllabification generalized to the voiceless plosives.
As for the remaining plosives, band d, different developments occurred. Earlier in the
Portuguese section 43 showed that $kl and $pl are more preferred syllable heads than are $df45 and
$bl. It also showed that d$1 and b$1 make better contacts than k$1 and p$/. It is probable then that
these voiced plosives remained in the coda position rather than becoming heads through resyl-
labification. This placed the voiced plosives in a position of weakening and this is what happened
to b and d. The syllable-final b weakened to a glide and colloquially d weakens to an r.
Finally, the remaining undesirable syllable contact of g$/ was again altered through
resyllabification. Thus the last stage in our derivation, as shown below, is slope steepening which
happened only with the velar plosives.
(49) dup$1u orik$1a stab$1u vig$11ire
Resyllabification I: du$plu ori$kla
Coda Weakening: stau$1u
Resyllabification II: vi$glar
Slope Steepening: ori$k,Ja vi$g.ia
duplu oreche46 staul veghca4 7
44 -ti- > -kl- .
45 Of course this type of syllable head is not possible in Romance, therefore coda weakening was a viable
alternative.
46
The palatalization of [k] and [g] does not appear to be an act of progressive assimilation; that is, these two
plosives were palatalized by the following yod (</)and not the preceding front vowel. This is shown by the
fact that examples such as Rom. mic 'small' (Lt. micu ), Rom.frig 'cold' (Lt.frigu) do not show palatalization
~also see Nandris: 146 & 150) .
7 While palatalization of I>}does take place when I is syllable-initial, e.g., Lt. Iinu, Rom. in 'flax', Lt. l~pore,
Rom. iepure 'rabbit, hare' (see Nandris: 140). this change only occurs when I is followed by i; therefore, it
appears that this is a case of assimilation and not a change based on consonantal strength.
58
 A comparison of Italian and Romanian syllabification shows that in Italian VT$rV >
V$TrV and VT$1V remained, that is, there was no resyllabification. In Romanian though
resyllabification generalized with the voiceless plosives and liquids, that is, VT$LV > V$TL V.

5.1.1 Affrication
Another bit of evidence which indicates heterosyllabification, or VC$CV, is the develop-
ment of -dJ- and -tJ-. These sequences had two different but parallel developments, and each
development was based on stress assignment and syllable structure. Word stress on the vowel
preceding the plosive tended to close the stressed syllable; that is, f d$J and ft$J were the result.
If however stress was on the vowel followin_g d and t this tended to attract d and t to the head
position of the stressed syllable; that is, $dJV and $UV' resulted (Nandris:l28). The outcome for
both tautosyllabification and heterosyllabification is affrication but the end products differ, as
shown in 50 and 51.

(50) '9'd$jV >dz $dj~> $f

'9't$jV > tS $U~ > $t

(51) Lt. mMiu Rom. miez (dz> z) 'core; kernel'

Lt. adjutare Rom. ajuta (dt > t) 'to help'

(cf. Rom. jos, Lt. deo(r)sum 'low; down')

Lt. negotiu Rom. negot (ls) 'trade'

Lt. +fetiolu Rom. fecior (t~) 'boy; son'

Thus it appears we have direct evidence for an earlier VC$CV syllabification, in the case
of stress on the first syllable.

5.1.2 Metathesis
The voiced labials b and m, as well as J: underwent another method of syllable contact
improvement.48 When these elements appeared in the coda position and were followed by yod,
the yod metathesized thereby producing a more preferred sequence of W$S rather than S$\V.

(52) Lt. scabia ORom. zgaibll '!

Lt. diffamiat Rom. deftiima 'defame'?
Lt. cofea Rom. coif 'helmet'

There are many processes which conspire to create a more preferred syllable structure.
The processes reviewed above, slope steepening, coda weakening, metathesis and varying
phonological developments based on different syllabifications, can all be explained when we
assume an earlier syllabification of VC$CV.

48 The only Romanian word available with a sequence of p +J available was apropia(re), Lt. appropiare 'to
draw, bring near' . The lack of any change suggests that the i in this form may have remained syllabic, thus
no change was required.

59
6.0 Catalan
6.1 Gemination
Catalan, like Italian, employed gemination as a means of improving a syllable contact of
VC$CV.49
(53) Lt. diabl:Slus Cat. diable [diabblOJ] ?
cf. Lt. nob(i)le Rous. [nobbfa po 'noble'
Lt. dilplus Cat. doble [dobbl1J1s1 'double'
Lt. dilplus Rous. [dubbliJ] 'double'
Lt. regulum Cat. regla [rreggfa1s2 'rule'
Lt. si!cille Cat. segle [segglOJ] 'century'
? Cat. tee la [tekklOJ )53 'key'
cf. Lt. +joc(u)lare Rous. [fuggh1] 'to joke'
As can be seen in the examples in 53, when a strong plosive in the coda position meets
a weaker liquid in the head position, gemination is the result. This gemination occurs in order to
satisfy the Syllable Contact Law. After gemination a strong coda and an equally strong head come
in contact, a situation preferable to the original.
Stress seems to play a factor in the syllabification and gemination process in Catalan.
Huber (1929:11) mentions that when the vowel preceding the plosive is stressed, gemination is
the result, but when stress follows the plosive, there is no gemination. The first situation, in which
stress precedes the plosive, would result in a closed syllable as stressed syllables tend to attract
segments to them. Thus we would have VC$CV. When the stress follows the plosive, the result
is a complex head in the stressed syllable; that is, V$CCf. This would explain the varying results
seen in d6ble [-bbl-] and doblar [dubl~] 'to double' (Huber:l 1). In doble the stress precedes the
word-internal plosive and gemination is the result. In doblar, where stress comes after the
intervocalic plosive, there is no gemination (cf. section 4.1.1.on Portuguese and 5.1.1 on
Romanian).S4
While the syllabification for an intervocalic plosive plus a lateral liquid was heterosyllabic,
that for a plosive plus r appears to be tautosyllabic based on the developments of the plosive;
namely intervocalic weakening.

49 ' he change of r > rr syllable-initially, as in Lt. rr1beus, Cat. roig [rr:Jt] , is a means ofonset strengthening
and not syllable contact induced gemination.
50 Rousillon is a dialect of Catalan.
51 Intervocalic voicing could have taken place prior to or after gemination. The result would be identical.
52 Cf. Cat. reglar [rragla] 'to rule' .
5 3 Cf. Cat. teclat [tlklat] 'keyboard' .
54
The non-geminate sequence in doblar remains controversial as the Diccionari ortografic i de pronuncia
(1990) does have [dubbla].

60
(54) Intervocalic weakening: VCV (where C represents any plosive)
Lt. in\lltare Cat. envidar [ambilSa)Ss 'to invite'
Lt. arripare Cat. arribar [aril3a]S6 'to arrive'
Lt. fricant Cat. fregan [frayan]S7 'they rub'
Lt. hlbernu- Cat. hivern [i&rn] 'winter'
Lt. nllda Cat. nua [nua]ss 'naked (f.sg.)'

VCrV
Lt. petra Cat. pedra [pelSra]59 'stone'
Lt. supra Cat. sobre [soBra] 'over, above'
Lt. +acrus Cat. agre [ayra] 'sour'
Lt. febre Cat. febre [fe13ra] 'fever'
Lt. quadr!ta Cat. cairat [kajral] 'beam'

Comparing the examples in 54 with those in 53 we can see that the type of development
was primarily based upon the syllable structure. When the plosive and liquid were heterosyllabic
(53), gemination took place, however, when the plosive and liquid were tautosyllabic (54),
intervocalic weakening occurred. Because the r is weaker than tfie /,and therefore makes a worse
contact, heterosyllabification was less tolerated and resyllabification occurred. A complex head
of $Cr (where C represents any plosive) is more preferred to $Cl according to the Head Law. thus
original C$l did not resyllabify. Instead, gemination took place which also resulted in a bctler
syllable structure.

6.1.1 Coda Weakening

Besides gemination, coda weakening was also utilized in Catalan as a syllable contact
improvement process. The SCL is satisfied by weakening a coda with a stronger consonantal
strength than the following syllable head.

(55) Lt. scop(u)lu Cat. escull [asku,(]6 'rock'

cf. Lt. scl5pillus Rous. [askui\] 'rock'
Lt. tabula Cat. taula [taula] 'table, board'
Lt. oculus Cat. ull [ui\] 'eye'
Lt. vetulus Cat. vell [ve,\)61 'old'
Lt. coagulu Cat. coall [koai\] 'dabber'

55 t > d > "

56 p > b>IJ.
57 k > g > y.
58 The transition was most likely d > "> a .
59 There are also forms such as Cat. mare 'mother', Lt. matre and Cat. pere 'father', Lt. patre, in which
t >">a and earlier forms with -pr-> -br- > ur. e.g. Lt. cupru, + cubru >Cat. coure ?[koura] 'copper' . As
well the Catalan word cairat also indicates coda weakening. These forms could indicate an earlier
heterosyllabic sequence which would explain the coda weakening.
60 -pl- > -kl- .
61 Recall that -ti- also became -kl- .

61
 . In the sequences-k/-(from original-kl-, -pl- and-t/-)and-g/- the stop weakened toayod,
palatalized the following I, as indicated by the [.<] in all but one of the examples above, and then
deleted. In tau/a the plosive also weakened to a glide but this had no effect on the following
consonant. The weakening of a plosive to a glide produced a more preferred sequence of W$S as
opposed to the original S$W sequence. The examples in 54 as compared to those in 55 once again
suggest that a difference in the syllable structure can account for the disparate developments.
The reader may be wondering why the same segments, p, b, t, k and g in the same
environments underwent different developments. It has been suggested (Fouche:l59) that in
learned words the process of gemination took place but in non-learned words coda weakening
occurred instead. Regardless of which development took place, both gemination and coda
weakening have the same objective: to satisfy the SCL.

6.1.2 Metathesis
There were several developments which took place when a consonant originally in the
coda position was followed by a yod. One of these developments was metathesis.

(56) Lt. basium Cat. bes [bes)62 'kiss'

Lt. bassiare Cat. baixar ~~a]63 'to lower'
Lt. corium Cat. cuir [latjr] 'leather'
Once the yod and plosive metathesized, due to the poor syllable contact, the yod
coalesced with the preceding vowel.

6.1.3 Onset Strengthening

When certain plosives preceded the yod, onset strengthening took place with subsequent
coda deletion.
(57) Lt. rubeu . Cat. roigJ.rut~] 'red'
Lt. podiu Cat. [put ] 'hillock'
With the sequences of d$J and b$J there seems to have been a strengthening first of the
yod, which can also be seen word-initially in Cat.jove ~Ula] 'young', Lt.jr1vt!nis. After the yod
was strengthened, the stop in the coda position was deleted.

(58) rub$Ju pod$J~

Glide Strengthening: rub~fu ~uu
Coda Deletion: ru~ u
~~ ~
Final Vowel Deletion:
Final Devoicing:
(Vowel Alteration): ru~ pu~

ru~ put~

62 s > z; Catalan has final devoicing.

63
Palatalization took place with -ss- .

62
 The development of p$J differs from the plosives above. In some cases it appears that
this sequence may have resyllabified, that is,p$J> $pJ, e.g., Lt. sepia, Cat. sepia [SEpia] 'cuttlefish'
. Since p has the greatest consonantal strength of any of the plosives and yod is one of the weakest
segments, it follows that p$J will make the worst contact. However, if resyllabification occurs $pj
is formed, resulting in the most preferred complex head. As we have repeatedly seen, change starts
with the worst structures; therefore, it is expected that if a sequence of a heterosyllabic consonant
plus yod is going to resyllabify, resyllabification will start with the least preferred.
In some other words it seems that i does not lose its syllabicity and thus a form like Lt.
sapiat remains trisyllabic, as in Cat. sapiga 'that he know' .64
This final branch of Romance shows many of the processes we have already seen in
previous sections of this paper, metathesis, coda weakening, glide strengthening and gemination.
In order to explain the developments systematically and economically, we need to begin with a
syllable structure that can account for all of the changes. The evidence points to VC$CV.

7.0 Proto-lndo-European Syllable Structure6s

The data from the four Romance languages above presents evidence for Proto-Romance
VC$CV. Evidence for this syllable structure also exists in Sanskrit, Proto-Germanic, Pre-Latin and
Ancient Greek. This means that VC$CV can likely be pushed back farther than just Proto-
Romance and into PIE. This is a very strong claim, but one not without support.

7.1 Germanic
Murray and Vennemann ( 1983) argued for VC$CV in Proto-Germanic with verification
for this argument coming partly from West Germanic gemination. Examples of this arc shown
below.

(59) Go. sat$jan OS settian OE settan 'to set'

Go. -skafi$jan OS skeppian OE scieppan 'to create'
Go. kun jis OS kunnies 'race (gen.)'
Go. hal$ja OS hellia 'hell'

Go. alas OS akkar 'acre'

ON ep$le OE reppel 'apple'

The data shows that before yod, all consonants except r geminated. This is due to the
fact that of all the consonants, r has one of the lowest consonantal strengths. Gemination started
with the poorest contacts and worked its way toward better contacts. The worst contact T$J has
a contact evaluation of -6 whereas r$J is only -1. Given r$j' s better evaluation, gemination did
not generalize to this sequence. Similarly, it was only the poorest sequence of T$1 (contact
evaluation -4) which geminated (cf. Italian 3.1 ). Once again gemination was utilized to improve
a sequence of VC$CV (S$W).

64 Thi! g in this word most likely arose through glide strengthening, e.g. pi$a > pi$Ja (with yod insertion)
> pi$ga (glide strengthening).
65 This section is based on my paper entitled Prolo-lndo-Europe<m Syllable Structure.
63
7 .1.1 Sanskrit
In Sanskrit verse there was a light/heavy syllable distinction. In a sequence of -VCCV-
the first consonant closed the first syllable, which means that the two consonants were heterosyllabic.
(60) Verse construction66
,;~~ Vf~e
VC$CV
Therefore, for the purposes of verse construction, there existed a heavy syllable type such
as VC$CV.
There were also accounts by ancient Indian grammarians that intervocalic consonant
clusters tended to be heterosyllabic. Thus a word like pitre would be syllabified as pit$re and not
pi$tre, and mukta- was syllabified as muk$ta not mu$kta (Varma 1961).
Gemination also occurred in Sanskrit. Forms such as addya < ad$ya 'today', puttrel)a
< put$rel)a 'son (instr.)' and prettya < pret$ya 'post-mortal state' (Varma 1961:62-64; from
Vennemann 1988:34) indicate that a poor syllable contact of a plosive plus a glide or a plosive plus
a liquid was ameliorated through gemination. As in West Germanic, gemination in Sanskrit
reflects a Ve $CV pattern.
7.1.2 Ancient Greek
Different Greek dialects show different syllabification patterns. In Homeric Greek we
seethe same type of syllabification evident in Sanskrit; that is -VCCV- was syllabified as VC$CV
(Murray 1988:213). However, in Attic Greek an intervocalic stop plus a liquid is tautosyllabic as
is a voiceless stop plus a nasal. Otherwise you still have VC$CV. This is represented below.
(61) Ancient Greek Syllable Structure67
H A1tk
VCSCV V$fLV,V$DLV
V$TNV
otherwise VC$CV
A piece of phonological evidence from Greek that reflects a pre-Greek pattern of VC$CV
is ro versus o in comparative and superlative adjectives ending in ot; In the comparative and
superlative adjectives ending in ot;, the vowel of the syllable following a heavy syllable would be
o. If however, the preceding syllable was light, the vowel of the following syllable would be ro
(Murray 1988:215; based on Hermann 1923).
(62) T$T Azu6tmo<; ( Ae7tt6<; 'thin, fine')
T$r aKQ,6tato<; ( mcp6<; 'long, big')
D$N .K~6tato<; (KEOV6<; 'careful')
D$1 &K7ta:Mtmo<; (&K1tayA6<; 'terrible')

66 Murray (1988:211).
67
Murray (1988:213).

64
 In the first example in 62 then we have to assume that-irr-, or-pt-, is heterosyllabic given
that the vowel following this cluster is o and not m. This is true of all of the underlined clusters
in 62.

7.1.3 Pre-Latin
. ~e have already s~n in the introduction to this.paper tha~ there was ~vidence ror yc$CV
m Pre-Lahn. Recall that mt~gra should have been int'1gra if the plosive and hqmd were
tautosyllabic, leaving the preceding vowel in an open syllable, as in inf1$cit. This indicates that
the syllable structure was VC$CV at a point earlier than Classical Latin. Evidence for
tautosyllabification exists in the ClassicaJ Latin words ten~$brae and plI$tris since in Classical
Latin a short vowel followed by a plosive and a liquid is considered to be in a light syllable.

7.1.4 Summary
Sanskrit, pre-Latin and pre-Greek all indicate a common syllable structure of VC$CV.
Changes of this sequence in Latin and Greek are due to poor syllable contact. In both Latin and
Greek we get tautosyllabification, although it has generalized further in Greek. Sanskrit remains
unchanged and is therefore considered archaic. These poor syllable contacts are reflected in the
consonantal strength evaluations in 64. The scale below differs somewhat from the one we saw
earlier. 68

(63) Consonantal Strength Scale

L N D T
1 2 3 3.5
(64) Structure Evaluation !J.. Qk.
T$L -2.5 + +
D$L -2 + +
T$N -1.5 +
D$N -1
( '+' indicates resyllabification)
The plus and minus signs show which language underwent resyllabification with a
particular heterosyllabic consonant cluster. The syllable structures beneath the word 'Structure'
in the diagram above are those posited for PIE.
The original PIE structures remained heterosyllabic in Sanskrit while there was
resyllabification in Latin and in Greek. What was considered tolerable in Sanskrit was not
tolerable in Latin and even less tolerable in Greek which again shows us the language-specific
variability in syllable structure tolerance.
Evidence from the various branches of Indo-European all reflect an earlier Proto-Indo-
European syllable structure of VC$CV. Gemination in West Germanic and Sanskrit can be
accounted for by poor syllable contact improvement as can the resyllabifications in Latin and
Greek. Vowel quality in Latin, as well as vowel quality in Gteek, and verse construction and

68 The scale in 63 and the evaluations in 64 are from Murray (1988:216-217).

65
ancient grammarian reports ii) Sanskrit support a reconstructed Proto-Indo-Europcan syllable
sequence of vacv.
8.0 Conclusion
I have used various developments from a wide spectrum of Romance languages (Italian,
Portuguese, Romanian and Catalan) to show that the ancestor of the Romance languages, Proto-
Romance, must have begun with a syllable structure of VC$CV and not V$CCV, as is commonly
believed. Due to the pressures applied to this original form, as outlined by the Preference Laws,
certain changes were forced. Improvement of the syllable structure can explain the historical
phonological processes which I have described above. These changes cannot be explained quite
so neatly, if at all, starting with Proto-Romance V$CCV. Thus the cumulative evidence of several
Romance languages confirms the argument for VC$CV in Proto-Romance. Evidence from other
Indo-European languages suggests that the syllable structure of Proto-Romance posited here was
inherited from Proto-Indo-European. A summary of the major changes in the Romance languages
I discussed is presented in Table 2.69
Table 2

OL VC$CV>
PR VC$CV> Italian Portmmese Romanian Catalan
Vp$jV Vp$pjV Vj$nV
Vt$jV Vt$tjV Vj$tV
(with (wih
valatalization) nalatalization
Vk$tjV Vk$1QV V$C
(complete
nalatalization)
Vb$jV Vb$bjV Vj$bV Vj$bV Vb$jV
(with onset
strengthening and
coda deletion)
Vd$jV Vd$cijV Vj$dV Vd$jV
(with (with (with onset
palatalization) palatalization) strengthening and
coda deletion)
Vg$gjV V$C
(with (complete
nalatalization) nalatalization)
Vv$rV V$nrV V$nrV V$orV
Vt$rV V$trV V$trV V$trV

69 Blank spaces in Table 2 indicate inconclusive evidence.

66
 '

Vk$rV V$krV V$krV V$krV

Vb$rV Vb$brV V$brV V$brV
Vd$rV V$drV V$drV Vd$rV
(with coda
weakenin2)
Vg$rV V$grV Vg$rV
(with coda
weakenin2)
Vp$1V Vp$plV Vp$1V/ V$plV Vp$plV/
(with slope V$plV Vp$1V
steepening) (with slope (with coda
steepening) weakening)
Vt$1V Vk$klV Vk$1V V$klV Vt$1V
(with slope (with coda (with slope (with coda
steeoening) weakening) steeoening) weakening)
Vk$1V Vk$klV Vk$1V V$klV Vk$klV/
(with slope (with coda (with slope Vk$1V
steepening) weakening) steepening) (with coda
weakenin2)
Vb$1V Vb$blV Vb$1V Vb$1V Vb$blV/
(with slope (with (with coda Vb$1V
steepening) assimilation) weakening) (with coda
weakening)
Vd$1V Vd$1V Vd$1V
(with (with coda
assimilation) weakening)
Vg$1V Vg$1V Vg$1V Vg$1V Vg$glV
(coda (with coda (with slope
weakening weakening) steepening)
and
assimilation)

67
 REFERENCES
Allen, W. Sidney. 1973. Accent and Rhythm. Cambridge: Cambridge University Press.

Bruguera i Talleda, Jordi. 1990. Diccionari ortogriific i de pro111incia. Barcelona: EnciclopCdia

Catalana.

Colomer, Jordi. 1973. Diccionari angles-catald, catala-angles. Barcelona: Llibrc de Butxaca.

Donegan, P. J. and D. Stampe. 1979. 'The syllable in phonological and prosodic structure'.
Syllables and Segments, ed. by A. Bell and J.B. Hooper, 25-34. Amsterdam: North
Holland.

DuNay, Andre. 1977. The Early Iii story ofthe Rumanian Language. Lake Bluff, Illinois: Jupiter
Press.

Elcock, W. D. 1960. The Romance Languages. London: Faber & Faber Limited.

Foley, James. 1970. 'Phonological distinctive features'. Folia Linguistica 4. 87-92.

_ _ _ . 1977. Foundations of Theoretical Phonology. Cambridge: Cambridge University

Press.

Fouche, Pierre. 1980. Phon~tique historique du roussillonnais. Geneva: Slatkine.

Grandgent, Charles H. 1927. From Latin to Italian. Cambridge: Harvard University Press.

Hall, Robert A. Jr. 1950. 'The reconstruction of Proto-Romance'. Language 26. 6-27.

1974. External History of the Romance Languages. New York: Elsevier.

1976. Proto-Romance Phonology. New York: Elsevier.

1983. Proto-Romance Morphology. Philadelphia: John Benjamins Publishing Co.

Halle, Morris and Jean-Roger Vergnaud. 1987. An Essay on Stress. Cambridge: MIT Press.

Hermann, Eduard. 1923. Silbenbildung im griechischen und in den andern indogermanischen

Sprachen. Gfittingen: Vandenhoek and Ruprecht. 2nd ed. 1978.
68
Hill, A.A. 1954. 'Juncture and syllable division in Latin'. Language 30. 439-447.

Hoenigswald, H.M. 1949. 'A note on Latin prosody: initial S impure after a short vowel'.
Transactions of the American Philological Association 80. 270-280.

Hooper, Joan B. 1976. An Introduction to Natural Generative Phonology. New York: Academic
Press.

Huber, Joseph. 1929. Katalanische Grammatik. Heidelberg: Carl Winters

Universitatsbuchhandlung.

Lindsay, W.M. 1894. The Latin Language. Oxford: Clarendon Press.

Meyer-Liibke, W. 1935. Romanisches erymologisches Worterbuch. vol. 1 & 2. Heidelberg: Carl

Winters Universitatsbuchhandlung.

Muller, H.F. 1929. A Chronology of Vulgar Latin. Tiibingen: Max Niemeyer Verlag.

Murray, Robert W. 1987. 'Preference laws and gradient change: selected developments in
Romance'. Canadian Journal of Linguistics 32(2). 115-132.

____ . 1988. Phonological Strength a11d Early Germanic Syllable Structure . Munich:
Wilhelm Fink Verlag.

____ . 1991. 'Early Germanic syllable structure revisited'. Diachro11ica VIII:2. 201-238.

Murray, Robert W. and Theo Vennemann. 1983. 'Sound change and syllable structure in
Germanic phonology'. Language 59. 514-528.

Nandris, Octave. 1963. Phonetique historique du rounwin. Paris: Librairie C. Klincksieck.

Nespor, Marina and Irene Vogel. 1986. Prosodic Phonology. Dordrecht: Foris Publications.

Newton, Brian. 1972. 'lnterdigitation in French phonology'. La11guage Sciences. February. 41-
43.

Nyman, M. 1982. 'Lexicalization out of casual speech: the Greek-Latin synizesis'. Relational
and Reconstructive Aspects of Grammatical Systematization. Data Oriented Studies.
Helsinki: University of Helsinki. 65-95.
69
 Panovf, Irina. New Pocket Romanian Dictionary. New York: Hippocrene Books, Inc.

Pei. Mario A. 1954. The Italian Language. New York: S.F. Vanni Publishers.

_ _ _ . 1976. The Story of Latin and the Romance Languages. New York: Harper & Row.

Pensado, Carmen. 1989. 'How do unnatural syllabifications arise? The case of consonant+ glide
in Vulgar Latin'. Folia Linguistica Historica VIII 1-2. 115-142.

Pulgram, Ernst. 1950. 'Spoken and written Latin'. Language 26. 458-466.

____ . 1975. Latin-Romance Phonology: Prosodics and Metrics. Munich: Wilhelm Fink
Verlag.

Richardson, Elbert L., Maria de Lourdes Sa Pereira and Milton Sa Pereira. 1943. McKay's
Modern Portuguese-English and English-Portuguese Dictionary. New York: David
McKay Company, Inc.

Rohlfs, Gerhard. 1966. Grammatica storica de/la lingua italiana e dei suoi dialetti. Vol. l. Turin.

Saltarelli, Mario. 1983. 'The mora unit in Italian phonology'. Folia Linguistica 17. 7-24.

Salverda de Grave, J. J. 1930. 'Sur I' evolution des consonnes en italien '. Romania 56. 321-330.

Taylor,James L. 1970. A Portuguese-English Dictionary. Stanford: Stanford University Press.

Varma. Siddheshwar. 1961. Critical Studies in the Phonetic Observations of Indian Grammar-
ians. Delhi: MunshiRamMonoharLal. [Reprintofthe 1929edition,London, TheRoyal
Asiatic Society.]

Venoemann, Theo. 1972. 'On the theory of syllabic phonology'. Linguistische Berichte 18. 1-
18.

____ . 1988. Preference Laws for Syllable Structure. Berlin: Mouton de Gruyter.

Williams, Edwin B. 1962. From Latin to Portuguese. Philadelphia: University of Pennsylvania

Press.

70
 APPENDIX
U1 voiced palatal central glide ~>[]in Romanian
[u] voiced .rounded labiovelar glide <t> [tS] in Romanian
[13) voiced bilabial fricative <x> []in Portuguese
[ti voiceless alveolar affricate C consonantal sound except where
[d ] voiced alveo-palatal affricate otherwise stated
[y] voiced velar fricative CL Classical Latin
[A] palatal lateral approximant D voiced plosive
fjl] voiced palatal nasal G glide
[OJ voiced interdental fricative L liquid
[t ] voiceless alveo-palatal affricate N nasal
~>[a] (between Mand /a:/) NRom. North Romanian, also known
in Romanian as Daco-Roman
<ila> closer and tenser than /a/ OL Old Latin
in Romanian PR Proto-Romance
<6> [o] in Portuguese T voiceless plosive except where
<c> [s] in Portuguese otherwise stated
<!;> [s] in Portuguese V vocalic sound
<ch> voiceless palatalized velar stop VL Vulgar Latin
when followed by <i/e> in Romanian V short vowel
<g> [z] when followed by <i/e> V long vowel
in Portuguese + reconstructed form when raised and
<gh> voiced palatalized velar stop when preceding a word
followed by <i/e> in Romanian + morpheme boundary
<j> [f] in Portuguese * hypothetical or incorrect form
<lh> [A] in Portuguese $ syllable division
<nh> fjl] in Portuguese >becomes
<derives from

71